This thesis presents a new method of the 2-D partially occluded object discrimination for the computer vision application. A binary modified Hopfield neural network was applied to perform the global feature matching. To obtain the feature points of the object, a Gaussian function was implemented to smooth the object boundary curve and a curvature estimation method was used to extract the dominant points. A 3-point matching method was used to perform the initial comparison and to build the disparity matrix. Finally, the coordinate transformation was used to eliminate the false matched points. Two image banks, a model object image bank and an occluded object image bank, were built for the discrimination test. The result showed that the discrimination algorithm was successful. |